To compare shoulder sonography (US) after intraarticular fluid injection ("arthrosonography") with conventional shoulder US for diagnosing anterior labral tears. Thirty consecutive patients who were to undergo shoulder MR arthrography were enrolled after written informed consent was obtained. A radiologist performed conventional US, using a 2- to 5-MHz convex transducer on the subject whose arm was externally rotated, then abducted perpendicular to the trunk, and rotated internally and externally. After intraarticular injection of diluted gadolinium contrast for MR arthrography, arthrosonography was performed in the same manner as in the conventional method. Two other radiologists compared the two US methods in the diagnosis of anterior labral tears using a five-point scale. Receiver operating characteristic (ROC) curves and weighted kappa values were calculated. Fourteen patients proven to have anterior labral tears by arthroscopy constituted the disease group, and 16 patients, including the 4 patients who were negative on arthroscopy, constituted the normal group. The areas under the ROC curve of arthrosonography (mean, 0.977) were larger than those of conventional US (mean, 0.858). The kappa values for assessing the interobserver agreement of the two modalities were 0.40 and 0.82, respectively. Arthrosonography of the shoulder is more accurate and reliable than conventional US for diagnosing anterior labral tear.